RIGHTS RESERVED. PLEASE DIRECT ALL REQUESTS FOR PERMISSION TO PHOTOCOPY OR REPRODUCE ARTICLE CONTENT THROUGH THE UNIVERSITY OF CALIFORNIA PRESS’S
RIGHTS AND PERMISSIONS WEBSITE, HTTP://WWW.UCPRESSJOURNALS.COM/REPRINTINFO.ASP. DOI:10.1525/MP.2009.27.1.61
The Chill Parameter 61
OLIVER GREWE, REINHARD KOPIEZ,AND ECKART ALTENMÜLLER
Hannover University of Music and Drama,Hannover, Germany
STRONG EMOTIONAL FEELINGS SEEM TO INCLUDE
physiological and motor responses, as well as the cogni-tive appraisal of a stimulus (Scherer, 2004). Individualsmay thus reach emotional peaks at different points intime. “Chills” (goose bumps or shivers) offer usefulindications of individual emotional peaks (Panksepp,1995). Reported chills of 95 participants in response toseven music pieces are presented. Subjective intensity aswell as physiological arousal—Skin ConductanceResponse (SCR) and Heart Rate (HR)—revealed peaksduring chill episodes. A possible influence of breathingwas excluded. Familiarity with the music had a signifi-cant impact on chills. Age, gender, and music educationshowed no influence on chill frequency. In anexploratory approach, the influence of active music lis-tening (i.e., singing along, lip syncing, etc.) could not beconfirmed. These results suggest that chills are a reliableparameter, synchronizing subjective feeling with thephysiological arousal component, without being influ-enced by motor responses.
Received September 2, 2008, accepted March 3, 2009.
THE RELATION BETWEEN THE THREE MAIN emo-tion components: subjective feelings, physiologi-cal arousal, and motor response (Scherer, 1994,
2004) is still under evaluation (Cacioppo, Klein,Berntson, & Hatfield, 1993; Krumhansl, 1997).Physiological reactions indicate bodily arousal, asreflected in changes in skin conductance, heart rate,and other physiological reactions. Motor reactionsinclude behavioral responses to a stimulus such as facialreactions, fight, or flight. Subjective feelings reflect theconscious evaluation of a reaction towards a stimulus.It is not clear whether these components always occur
in synchrony and whether each emotional reactionalways contains all three components. Furthermore, adifference between utilitarian and aesthetic emotionshas been suggested (Scherer, 2004). Emotions inresponse to aesthetic stimuli such as music might differfrom “real life” emotions with biological relevance. Inorder to make valid statements about subjective feel-ings, psychologists must rely upon participants’descriptions of what they perceive (Russell, 1997).Physiological measurements offer objective data.However, it is difficult to interpret their meaning interms of an emotional response. For the study of emo-tion, a parameter that combines subjective and objec-tive emotion components would be extremely useful.Furthermore, a simple triggering of strong emotionalresponses is difficult because strong feelings in differentindividuals might be elicited at different points in time.However, a promising indicator of individual peaksincluding different components of emotion are “chills.”Chills are bodily responses, such as shivers or goosebumps, that are elicited by mostly aesthetic stimuli(e.g., music) and are perceived as highly pleasurable(Panksepp, 1995, 1998; Sloboda, 1991). They mightserve as a useful parameter for emotion researchbecause they seem to involve physiological as well assubjective feeling aspects of emotion, and they can beregularly elicited by music (Sloboda, 1991).Additionally, McCrae (2007) found chills to be one ofthe best definers of openness to experience, which is oneof the five basic personality factors.
That chills in response to music are mainly perceivedas highly pleasurable occurrences can be concludedfrom several previous studies. Goldstein (1980) report-ed that between 76 and 91% of listeners labelled chills asbeing pleasurable, some participants even calling thesensation orgasmic. Furthermore, Goldstein found thatthe chill frequency was diminished in participants treat-ed with naloxone, a specific endorphin antagonist. Heconcluded that chills reflect “a spreading electrical activ-ity in some brain area . . . with neural links to the limbicsystem and to central autonomic regulation” (Goldstein,1980, p. 128). Furthermore, Goldstein highlighted therelation between chills and the euphorigenic effects ofthe opioid peptides (endorphins). A study by Blood and
THE CHILL PARAMETER: GOOSE BUMPS AND SHIVERS AS PROMISING
MEASURES IN EMOTION RESEARCH
Music2701_05 8/13/09 2:56 PM Page 61
Zatorre (2001) found that chills activated specific brainregions that commonly are associated with hedonicimpact, reward learning, and motivation. Their studyrevealed that the intensity of chills positively correlatedwith regional cerebral blood flow (rCBF) in the left ven-tral striatum and dorsomedial midbrain, and a decreaseof rCBF in the right and left amygdale, left hippocam-pus, and ventro-medial prefrontal cortex. These regionsalso were activated during consummation of illicitdrugs in humans (Breiter et al., 1997) and rats (Bardo,1998), as well as during food intake and sex in animals(Pfaus, Damsma, Wenkstern, & Fibiger, 1995).
So far chills have been used in several studies con-cerned with emotional reactions to music. Sloboda(1991) found that chills seem to be related to musicalevents, such as unexpected harmonies and suddendynamic and textural changes. Panksepp (1995) high-lighted the importance of being familiar with a certainpiece to perceive chills. He reported that chills aremore frequently perceived by females and in responseto sad musical pieces. Both studies, however, relyexclusively on participants’ reports of chills; reportedchills were not controlled by or compared to physio-logical measurements.
Several more recent studies included physiologicalmeasurements in response to music. Craig (2005)measured the Skin Conductance Level (SCL) and skintemperature of 32 students during chill episodes. MeanSCL was significantly higher during chill sections. Skintemperature showed no differences between chill andnon-chill episodes. Rickard (2004) reported that emo-tional music is the most powerful elicitor of chills com-pared to relaxing music, arousing music, and emotionalfilm excerpts. Skin Conductance Response (SCR) alsowas higher for emotional music compared to the othercategories of stimuli. Rickard (2004) suggested thatSCR and chills might be the best indicators of strongemotional responses. Guhn, Hamm, and Zentner(2007) reported physiological and chill reactions inresponse to three excerpts of classical music over time.They found distinct musical structures related to chillresponses. Furthermore, an increase of skin conduc-tance was reported during musical passages that alsoinduced more chills. The results also demonstrated thepossible relation of volume, physiological reactions,and chills. Blood and Zatorre (2001) found neitherchanges in SCL nor in skin temperature. However,heart rate (HR), electromyogram, and respiration wereraised significantly during chills.
To summarize these findings the following could behypothesized: (a) Chills are mostly perceived as pleasur-able affective events up to orgasmic sensations; (b) chills
have measurable physiological and central nervous cor-relates; (c) chills occur regularly in response to musicand may be related to distinct events. All of these find-ings make chills a promising parameter for studyingemotional reaction because two main emotion compo-nents (subjective feeling and physiological arousal) arecombined in a single event that can be elicited regularlyin participants using music as a stimulus.
However, there still are several difficulties concerningchills as a parameter for emotions. Most studies usingchills called their results preliminary or explorative.Concerning the physiological correlates of chills, therehave been contradictory results, and the influence ofbreathing on skin conductance and heart rate has notyet been controlled for (see for example, Berntson,Cacioppo, & Quigley, 1993). Additionally, it is stillunclear whether chills are a consistent, adequateparameter for all ages, both genders, and participantswith different levels of music education. Since emo-tions develop over time, we thought a temporal analy-sis of chills would be interesting, comparing the onsetof reported chills with changes in subjective and physi-ological parameters. The induction of the chill reactionpresented a problem as a research tool for emotion. Tocollect a reasonable number of chills, most experi-menters have relied on participant-selected pieces(Blood & Zatorre, 2001; Panksepp, 1995). Up to now, ithas been difficult to find a musical stimulus that reli-ably elicits chills in different participants. Thus, it istricky to compare the results of different studiesbecause the acoustical features of music (e.g., loudness)also may influence the physiological response. Konecni,Wanic, and Brown’s (2007) effort to prime chill reac-tions using national anthems, stories, architecturalobjects, or paintings and thus increase the number ofchills also was not successful. The familiarity with apiece of music seems to have an impact on the frequencyof chills (Panksepp, 1995). For chills to become anestablished tool for emotion research, there needs to bemore clarity about how familiar exactly the participantsshould be with the music and how the frequency ofchills can be optimized in a heterogeneous group of lis-teners listening to the same piece of music.
Based on a previous experiment (Grewe, Nagel,Kopiez, & Altenmüller, 2007) the experimental setting ofthis study was optimized to induce chills more reliably.Furthermore, different levels of familiarity were com-pared according to their impact on chill frequency.Finally, possible influences or interferences of the thirdemotion component—motor response—with chillswere of interest. This aspect was addressed in anexploratory approach in which the motor activity was
62 Oliver Grewe, Reinhard Kopiez, & Eckart Altenmüller
Music2701_05 8/13/09 2:56 PM Page 62
manipulated while listening. The basic idea of this studywas to use chills as an indicator of individual emotionalpeaks. Participants may reach their emotional peak atdifferent points in time and in response to differentmusical events. The underlying physiological and psy-chological principles may be the same, though elicitedby different events. The underlying assumption is thatchills are not elicited by a simple stimulus-responsemechanism but that they are based on cognitive evalua-tion (Grewe et al., 2007). If chills are based on an indi-vidual appraisal process (see Ellsworth & Scherer, 2003),they would reflect more complex emotional reactions or“refined emotions” (Frijda & Sundararajan, 2007). Thisapproach is based on the individual appraisal of music,in contrast to other approaches that formulated the“production rules” for emotions in response to music(e.g., Scherer & Zentner, 2001). When complex emo-tions involving cognitive evaluation cannot be “trig-gered” in a simple manner, an objective indicator isneeded for individually occurring responses. The aimsof this study were to understand whether chills could besuch an indicator and to obtain answers to the followingquestions: (1) Are chills independent of the age, gender,and music education of participants, and thus can beused as a valid parameter? (2) Are chills perceived asspecifically intense, and is this intensity of feelingreflected in a measurable physiological arousal? If so,does the perceived intensity of subjective feelings andphysiological arousal reveal a characteristic relation overtime during chill episodes? (3) Is a general familiaritywith a piece of music sufficient for a reliable inductionof chills, or can the frequency of chills further beincreased by an increasing knowledge of a piece (culmi-nating in an intimate familiarity with a distinct record-ing)? (4) Does a motor response, such as singing along,have an impact on chill frequency?
Method
Participants
In this experiment 95 participants were tested (50females and 45 males, mean age = 40, SD = 16, Range =19-75). Participants were asked to estimate their level ofmusic education. A total of 36 reported being nonmusi-cians while 30 participants rated themselves as hobbymusicians; 12 were lay musicians with regular music les-sons, and 17 were professional musicians. The generaleducation and social backgrounds were heterogeneous:The group contained pupils, teachers, housewives, den-tists, retired people, and engineers, among others. Allparticipants sung in nonprofessional choirs.
EXPERIMENTAL GROUP
A subgroup of 54 participants had performed theMozart Requiem KV 626 in public concerts and thuswere highly familiar with the music (experimentalgroup). In order to test for familiarity effects, we subdi-vided the experimental group into three subgroups:choir A (N = 19), choir B (N = 14), and choir C (N = 21)(see Table 1). This subdivision was relevant for thedetailed analysis of the effects of familiarity. Recordingsof Choir A and B were used as stimuli. Two interpreta-tions of the Lacrimosa, Rex Tremendae, and Confutatisfrom Mozart’s Requiem were used in this experiment,one sung by choir A (version A) and one sung by choirB (version B). Additionally, we used the motet, UnserLeben ist ein Schatten [Our life is a shadow] by J. S.Bach, sung by choir A, and the Requiem by G. Puccini,sung by choir B. Choir C sung the Requiem KV 626 byMozart; however, no recording was used from choir Cin this experiment. Choir C was not familiar with theBach motet or the Puccini Requiem. The set of stimuliand the logic of comparisons are described in moredetail below.
CONTROL GROUP
The remaining 41 participants (control group)reported knowing none of the pieces. After listeningto each piece they were asked how well they knewthis particular piece. They reported a mean of 1.8(SD = 1.4) on a 7-point Scale (“1” = “not at all” and“7” = “very well”). These 41 participants sung in popand gospel choirs and, for the most part, were notvery familiar with classical music. Thus, this groupwas considered as being completely unfamiliar withall the stimuli. All participants were contacted dur-ing rehearsals and were not paid for their participa-tion in the experiment. The experimental group andcontrol group were balanced for gender, age, andmusic education.
The Chill Parameter 63
TABLE 1. Overview of Familiarity Levels.
Experimental Control Group Group
N 54 41
Subgroups Choir A Choir B Choir C
N 19 14 21 41Sung Requiem + + + −
in concertRecordings of + + − −
choir used in (Version A) (Version B)experiment
Music2701_05 8/13/09 2:56 PM Page 63
Musical Stimuli
The musical pieces for this experiment were selectedto obtain a maximum number of chills. Since famil-iarity was found to influence the frequency of chills(e.g., Panksepp, 1995), most of the stimuli that wereused in this experiment were performed by subgroupsof the participants (see Table 1 and Table 2).Movements from Mozart’s Requiem KV 622 were usedas stimuli as well as the motet, Unser Leben ist einSchatten [Our life is a shadow], by J. S. Bach andRequiem by G. Puccini. The set of stimuli containedprofessional recordings by Karajan (1989). These werethe Tuba Mirum and Dies Irae from Mozart’s Requiem.Additionally, it contained non-professional recordingsperformed by the participating choirs. These were theLacrimosa, Confutatis, and Rex Tremendae fromMozart’s Requiem, as well as the Bach motet and thePuccini Requiem. All pieces were played twice to theparticipants: the Tuba Mirum, Bach motet, andPuccini’s Requiem each in one version; the Lacrimosa,Confutatis, and Rex Tremendae each in two differentversions. Stimuli were presented in pseudo-random-ized order. All pieces were played once before theywere repeated. The design of this study allowed com-parisons on different levels:
COMPARISON OF FAMILIARITY WITH THE STIMULUS
We created three levels of familiarity on which to testthe participants. The control group was not familiarwith the musical stimuli, representing the lowest levelof familiarity. Choir C was familiar with Mozarts’Requiem. For choirs A and B we included some of theirown recordings, which represented the highest level offamiliarity. It must be considered that familiarity isalways related to preference in this context, since thestimuli are real musical pieces.
COMPARISON OF INDIVIDUAL INTERPRETATION VS. DIFFERENT INTERPRETATION
The selection of stimuli allowed us to examine whetherthere is a strong emotional impact of a “subjective”
identification effect with the individual interpretationof a musical piece. For example, a stronger emotionalreaction of choir A to version A of the Lacrimosa wasexpected, whereas for choir B a stronger reaction to ver-sion B was hypothesized.
COMPARISON OF REPETITIONS OF THE SAME STIMULUS
The repetition of the Bach motet, Puccini Requiem,and Tuba Mirum enabled us to test for the stability ofreactions.
Questionnaires
Participants filled in a short questionnaire prior to theexperimental session, reporting their musical knowl-edge, preferences, and demographic data. After listen-ing to each piece, they filled in a second questionnaireregarding the feelings they had during the listening ofthe piece. Feelings were reported on 7-point scales ofvalence and arousal.
Apparatus
SELF-MONITORING
Participants were asked to report the intensity of the affec-tive reactions they perceived continually while listening toeach piece. For continuous measurement of self-reportedaffective reactions, the EMuJoy software was developed(Nagel, Kopiez, Grewe, & Altenmüller, 2007). Participantsmoved a cursor on a computer screen in front of them inorder to express the intensity of their feelings. The cursorcould be moved continuously between two extremes. Theparticipants were instructed to move the cursor to the bot-tom of the screen if they did not perceive any affectivereaction at all and to move the cursor to the top of thescreen if they experienced extremely intense feelings. Datafrom EMuJoy were synchronized with physiological andmusical data in the range of milliseconds.
SELF REPORTS OF CHILLS
Participants were asked to press a mouse buttonwhenever they perceived a chill—defined as having a
64 Oliver Grewe, Reinhard Kopiez, & Eckart Altenmüller
TABLE 2. Overview of Musical Stimuli.
Bach Puccini Mozart Requiem motet * Requiem
Lacrimosa Confutatis Rex Tremendae Tuba Mirum Dies Irae
Recording of choir A B A B A B Professional Recordings A B(Karajan, 1989)
Note: * Title: “Unser Leben ist ein Schatten” [Our life is a shadow].
Music2701_05 8/13/09 2:56 PM Page 64
goose-bumps reaction (Gänsehaut) or experiencingshivers down the spine (Schauer über den Rücken)—while listening to the musical pieces. They were asked topress the button as long as the chill lasted. Participantschose which hand they preferred to use the mouse.Several pre-experiments as well as a previous study(Grewe et al., 2007) showed that the mere pressing ofthe mouse button did not lead to strong physiologicalreactions.
TECHNICAL DEVICES
Participants listened to the music via dynamic studioheadphones (Beyerdynamic DT 770 PRO) and a USBsoundcard (Audiophile, M-Audio). Physiologicalmeasurements were recorded with ARBO Ag/AgCl-electrodes (15 mm diameter) and amplified 100 timeswith a biosignal amplifier developed by the Institutefor Explorative Data Analysis (Institut für experi-mentelle Datenanalyse, IED), Hamburg. The analo-gous data from the EMuJoy program were digitalizedusing an A/D converter card (DT 301; DataTranslation, Marlboro, Massachusetts, USA). Heartrate was recorded with a heart belt (POLAR T31;Polar Electro Inc, Lake Success, USA). The analogousoutput from a receiver module (NRM receiver mod-ule; Polar Electro Inc, Lake Success, USA) was used tocalculate heart rate. The output consisted of a trail ofpulses that were converted into a time series of inter-beat-intervals. HR was calculated from the firstrecorded heartbeat. Time series of inter-beat-intervalswere interpolated using cubic interpolation to computeHR. Breathing was recorded with a respiratory effortsystem containing a piezo sensor (Item 1310, SleepmateTechnologies, Midlothian, VA, USA). Participantswrapped the belt around their lower thoraxes, so thatthe extension of the stomach due to breathing could bemeasured.
The physiological data, music, and EMuJoy datawere synchronized and recorded using the researcher-developed software based on DT Measure Foundry(Data translation, Marlboro, Massachusetts). For thetime series analysis, physiological data were sampledfrom 10,000 Hz down to 100 Hz after low pass filter-ing using the signal processing toolbox from MatLab.The down-sampling was executed with the decimatecommand (MatLab), which uses a Chebyshev Type Ifilter with a normalized cut off frequency of 64 Hzand 0.05 dB of pass band ripple. For EMuJoy, data fil-tering was omitted to avoid modifying narrow edgesof the self-report data. For data analysis, the programsMatLab (Version 7.1), Adobe Audition (Version 1.0),dBSonic (Version 4.13), and SPSS (Version 13.0)were used.
Procedure
All experiments were performed in individual sessions,to guarantee that participants could concentrate on themusic, their own feelings, and the rating task. The par-ticipants were introduced to the experimental task in astandardized manner after they had filled in the demo-graphic questionnaire. During the setting of the elec-trodes, the principles and devices used for physiologicalmeasurements were explained. Then a five-minutebaseline was taken while the participant was asked torelax in silence. After finishing the baseline, participantswere asked to breathe in deeply. During a recording ofapproximately one minute, this process was repeatedthree times. The recording was used to test for the influ-ence of breathing on skin conductance and heart rate.Then, participants were introduced to the EMuJoy andexplained the concept of “intensity of feelings.” To givethe participants the opportunity to test the EMuJoy andto try out the self-monitoring, we asked them to ratethe intensity of their feeling responses to 10 selectedpictures from the IAPS (Lang, Bradley, & Cuthbert,2001) before beginning the experiment. The followingpictures were presented in a fixed order for all partici-pants: a rafting scene (IAPS No. 8370), a graveyard(9220), a slit throat (3071), a tiny rabbit (1610), a spoon(7004), a child (9070), a female in an erotic pose (4220),a scene in a hospital (2205), a scene of violence (3530),and a basket (7010).
After asking the participant whether he or she under-stood the use of the EMuJoy, the experiment began.During the experiment, participants relaxed in a com-fortable armchair and listened to the music via closedheadphones. The participant and experimenter wereseparated by a room divider. The five music pieces werepresented in randomized order. The duration of theexperiment was about one and a half hours.
Further Data Processing
SKIN CONDUCTANCE DATA
When recording the Skin Conductance Level (SCL,tonic part), we measured the absolute values of indi-vidual skin conductance, which depend on many fac-tors, such as the moisture level of the skin, temperature,and blood flow. As a standard procedure, the high passfiltered SCL, the Skin Conductance Response (SCR,phase part) is calculated (Boucsein, 2001).
COMPARING CHILL SAMPLES AND NON-CHILL SAMPLES
In order to compare the self report and physiologicalchill data, we collected 622 chill samples from all data.The term chill sample refers to the data collection from
The Chill Parameter 65
Music2701_05 8/13/09 2:56 PM Page 65
the self-reported intensity of feelings as well as the SCR,SCL, HR, and breathing sample for that particular chill.The idea behind the collection of chill experiences wasthat strong individual emotional reactions may differ asto the point in time when they occur. Previous studieshave shown that chills cannot be “triggered” in a simplestimulus-response manner (Grewe et al., 2007; Guhn etal., 2007). Thus, the perception of a chill was used as anindicator for individual emotional peaks, independentof the distinct musical event that elicited the chill. Foreach chill, the corresponding events in reported inten-sity of feelings, SCR, SCL, HR, and breathing were cutout in 20 s windows, beginning 10 s before the chillonset. For each chill, a corresponding non-chill samplefrom the self report and physiological data of the sameparticipant was cut out. These non-chill samples wererandomly collected from periods of the same piece inwhich no chill was reported. Chills occurring in the first10 s of a piece were not collected because the physio-logical data would have been influenced by the orienta-tion response (Sokolov, 1990) due to the beginning ofthe stimulus.
Since participants reacted with different numbers ofchill, the mean of all chill excerpts for each participantwas calculated. Thus, each participant mean includedthe average subjective intensity of feelings and physio-logical chill response to the analysis. Means of chillsamples were compared with non-chill samples. Dataof HR and SCL were normalized according to theiraverage starting point to avoid influences of the indi-vidual variety in heart rate and skin conductance level.To test for the significance of differences between thesamples, a permutation test with 5,000 random per-mutations was performed for each set of data (Good,1994). This non-parametric test compares two datasets based on two matrices. For each permutation, ele-ments of the two matrices are randomly distributed ontwo new matrices. For instance, if a matrix of chillsconsists of the datasets A1, A2, A3 . . . and the matrixof the non-chill control consists of the datasets B1, B2,B3. . . , a possible random permutation would be B4,A7, B8 . . . vs. A3, A9, B11 . . . The test compares thedifferences between the averages of the chill and non-chill matrix. When less than 5% of the random per-mutations result in a bigger difference compared to thedifference between the two original groups of datapoints, the test becomes significant. To estimate thestrength of effect the Cohen’s d value was calculated forthe most extreme difference between both curves. ACohen’s d of 0.2 is interpreted as a weak effect, a valueof 0.5 as a medium effect, and a value of 0.8 and higheras a strong effect.
Explorative Comparison of“Active” and “Passive” Listening
In an exploratory approach, the Dies Irae from theMozart Requiem was compared in two different listen-ing conditions. In the active condition participantswere asked to sing along with the choir (or only to lipsync, if they preferred) and to imagine being a part ofthe choir, participating in the performance. In the pas-sive condition, the participants were asked to listensilently and imagine being part of the audience. For thisexperiment we used a professional recording (Karajan,1989; see Table 2). The physiological data from thisexperiment was not included in the analysis of chillphysiology since movement heavily influences thephysiological measurements. We compared the averagenumber of reported chills and the ratings of valenceand arousal given after listening to the piece. Active andpassive conditions were included.
Control Experiment: The Influence of Deep Breathing on Physiological Responses
The influence of breathing on skin conductance andheart rate is known as sinus arythmia (Berntson et al.,1993). As mentioned above, participants were asked tobreathe in deeply while their skin conductance andheart rate were recorded. The maximum of breathingfrom each recording was selected and an excerpt of 20 swas cut out of the recordings of breathing, SCR, andHR, spanning from 10 s prior to 10 s after the maxi-mum of breathing. The maximum of breathing wasused to interpret the values of breathing during the col-lected chill episodes.
Results
General Frequency of Chills
About one third of our participants (33) did not reportany chill. For the experimental group, 72% of the 54participants reported having at least one chill duringthe experiment, whereas only 56% of the 41 controlparticipants reported a minimum of one chill. A totalof 852 chills were collected (679 from the experimentalgroup, 173 from the control group). The maximum ofreported chills by a single participant was 88 for thewhole experiment. On average, each participant experi-enced 9 chills (SD = 16) during the experiment; however,the median was only 2 chills (upper quartile = 10, lowerquartile = 0).
66 Oliver Grewe, Reinhard Kopiez, & Eckart Altenmüller
Music2701_05 8/13/09 2:56 PM Page 66
Chills Showed No Relation to Age,Gender, or Music Education
A correlation between the absolute number of chillsreported and the age of the participant was not signif-icant, rS(93) = −.02, p = .25, N = 95). There was also nosignificant difference between the number of chillsreported by female and male participants, U(1) =1070.00, Z = −.42, p = .68). Females reported 9.2 chillson average (SD = 15.8, median = 2.5, upper quartile =0, lower quartile = 10.3) and male participants 8.7chills (SD = 16.4, median = 1, upper quartile = 0, lowerquartile = 10.5). A Kruskal-Wallis test revealed thatthe music education of the participants had no influ-ence on the number of chills they perceived, χ2(95) =2.40, p = .49.
We repeated all tests comparing only a selection ofour participants. Performing Mann-Whitney U testsbetween non-chill responders (33) and high chillresponders (33) who reported more than 5 chillsduring the experiment, we again found no significantdifferences based on age, gender, or music education.There was a difference between chill responders andnon-chill responders regarding being familiar withand liking classical music: liking of classical music:U(1) = 397.50, Z = −1.99, p < .05; being familiar withclassical music, U(1) = 386.50, Z = −2.07, p < .05. Chillresponders rated their liking of classical music at 6.2on average (SD = 1.2, median = 7, upper quartile = 7,lower quartile = 5) on a 1 to 7 scale. Non-chill respon-ders, in contrast, reported their liking of classicalmusic at 5.7 on average (SD = 1.1, median = 6, upperquartile = 7, lower quartile = 5). Chill respondersrated their general familiarity with classical music at4.8 on average (SD = 1.3, median = 5, upper quartile = 6,lower quartile = 4), non-chill responders at 4 on aver-age (SD = 1.5, median = 4, upper quartile = 5, lowerquartile = 3). However, there was no correlation acrossall participants between the number of chills and theknowing and liking of classical music. For other musi-cal styles, such as jazz, pop, and rock music, no rela-tions were found.
Example of Emotional Reactions to Music
The usual method to evaluate emotional reactions isto calculate the averages of ratings or physiologicalresponses. Figure 1 shows one example of such anevaluation. The experimental group and controlgroup are compared in their reactions to Puccini’sRequiem. This example was chosen to demonstrate
that an individual evaluation would be more benefi-cial for the understanding of emotion. Regarding thereports on intensity of feelings, the feelings of bothgroups were similar on average. However, physiologi-cal responses were stronger for the control group.Chill responses occur at different points in time andshow no clear relation to the average intensity reportor the averaged physiological response. The chillsindicate, however, that individual listeners reachedpeaks in response to the music at different points intime. Therefore, we collected and evaluated suchindividual events instead of presenting averageresponses. Exactly what it was in the music that elicitedsuch strong reactions is not of so much interest to usbecause these features seem very individual (for suchan analysis see, e.g., Sloboda, 1991, and Guhn et al.,2007). Rather, our focus is on the information gainedby the patterns showed by a strong emotional reac-tion. Chills seem to help pinpoint these strong emo-tional reactions.
The Chill Parameter 67
FIGURE 1. Example of emotional reactions to an excerpt of Puccini’sRequiem, symbolized by the acoustical envelope. Reactions of experi-mental group indicated by arrows. Ratings of intensity of feelings on ascale of −10 to 10. Chills in absolute numbers. SCR in µS.
Music2701_05 8/13/09 2:56 PM Page 67
Subjective Feeling Component of Chills
CHILLS SHOW A PEAK IN SUBJECTIVE INTENSITY
We asked participants to rate the intensity of their feel-ings over time while listening to the pieces. The ratingsfor the collected chill samples are shown in Figure 2.The intensity of feelings started rising several secondsbefore the chill onset. About one second before partic-ipants pressed the chill button, the intensity ratingsbecame significantly higher compared to random non-chill samples. After the chill onset, the intensity of feel-ings did not increase much further. To control theeffect size, we compared the most extreme difference(occurring at 6.16 s after chill onset) between bothcurves. Cohen’s d at this point was 4.44, which indi-cates a strong effect.
CHILL PIECES ARE RATED HIGHER IN VALENCE AND AROUSAL
Additionally, participants rated their feelings on theaxes valence and arousal after listening to each piece.The ratings for those pieces that elicited chills differedfrom those that did not elicit chills in both valence andarousal: valence, U(1) = 65599.00, Z = −9.69, p < .001,arousal, U(1) = 7817.00, Z = −6.82, p < .001. Chill pieceswere rated an average valence of 5.6 (SD = 1.4, median =6, upper quartile = 7, lower quartile = 5, N = 243) andan average arousal of 4.6 (SD = 1.9, median = 5, upperquartile = 6, lower quartile = 3) on a 1 to 7 scale. Non-chill pieces had an average rating of 4.6 for valence (SD =1.5, median = 5, upper quartile = 6, lower quartile = 4)and 3.8 for arousal (SD = 1.6, median = 4, upper quar-tile = 5, lower quartile = 3).
Chills Show Physiological Correlates in Skin Conductance and Heart Rate
As can be seen from Figure 3, breathing in has a mas-sive impact on physiological data. Before the experi-ment started we asked all participants to take a deepbreath, which we recorded. The synchronized record-ings provided two pieces of information. First, theyshowed the impact of breathing and second, theyhelped interpret the data. Breathing was recorded inarbitrary units, representing the extension of the lowerthorax of each participant. The synchronized, deepbreaths represent a maximum value for our group ofparticipants.
The physiological reactions during chills are repre-sented in Figure 4. Skin conductance level (SCL) showedan increase starting 2 s before the chill was reported bya button press. At 4.2 s after the chill onset it reached itspeak (Cohen’s d = 0.53). The skin conductance response(SCR) reflects the high pass filtered SCL, the phase partof the signal. The maximum increase in skin conduc-tance was reached 2.7 s after the chill onset (Cohen’s d =1.27). Interestingly, the increase in skin conductancebecame significant at the same time as the increase inthe intensity ratings (see Figure 1).
Heart rate also showed significant differencesbetween chill samples and random samples. One sec-ond after the chill onset the average HR was 1.5 beats
68 Oliver Grewe, Reinhard Kopiez, & Eckart Altenmüller
FIGURE 2. Ratings of subjectivly perceived intensity of feelings.Comparison of chill samples of 20 s length (arrow head) and randomsamples of the same length (arrow). Significant differences (permuta-tion test, p < .05) are grey shaded. The continuous intensity scale rangedfrom −10 to 10.
FIGURE 3. Reactions in Skin Conductance Response (SCR) and HeartRate (HR) to deep breathing. Data samples of 10 s length were matchedaccording to the maximum in breathing (Time 0) of all participants.
Music2701_05 8/13/09 2:56 PM Page 68
per minute faster compared to random samples.However, this difference became significant for a periodof only two seconds after the chill onset and reached itsmaximum at 5.3 s after the chill onset (Cohen’s d =0.58). The measurement of breaths showed a short last-ing significant difference 1.1 s before the chill onset.The difference between chill samples and random sam-ples at that point reached 0.5 units, Cohen’s d was 0.33,indicating a weak effect.
Familiarity Influences Chill Frequency
Within the groups of participants, several subgroupswere distinguished according to their familiarity withthe stimuli. We compared the average chill frequency foreach subgroup and piece (see Tables 2 and 3). Table 3shows the chill frequency in response to the three move-ments of Mozart’s Requiem that was played in two ver-sions. The number of chill responses showed significant
The Chill Parameter 69
FIGURE 4. Comparison of means of chill samples of 20 s length (arrow head) and means of random excerpts of the same length (arrows). The curvesrepresent Skin Conductance Response (SCR), Skin Conductance Level (SCL), Heart Rate (HR), and Breathing (Breath). Significant differences (per-mutation test, p < .05) are grey shaded.
TABLE 3. Comparison of Averaged Number of Chills for Subgroups with Different Levels of Familiarity.
Choir C (N = 21) 1.43f 1.05 f 1.10f 0.67f 0.62f 0.76f
Control group (N = 41) 0.61 0.34 0.15 0.02 0.24 * 0.02 *↑ ↑
χ2(3, N = 95) = 16.75 9.34 p < .01 p < .05
Note: f = familiar with piece, o = own recording (printed in bold), * = Wilcoxon test comparing the two versions A and B, p < .05.
Music2701_05 8/13/09 2:56 PM Page 69
differences between the groups for the Version B of theConfutatis, Kruskal-Wallis test, χ2(95) =16.75, p < .01,and the Version B “Rex Tremendae, Kruskal-Wallis test,χ2(95) = 9.34, p < .05. When comparing the differencesfor the two versions within the groups, only two casesshowed significantly different chill frequencies. Choir Areported 1.05 chills on average in response to its ownversion of the Lacrimosa, whereas the other versionelicited 0.95 chills on average, Wilcoxon Z(1) = −2.23,p < .05. In the group of participants unfamiliar with theMozart Requiem, reaction contained 0.24 chills on aver-age compared with version A of the Rex Tremendae and0.02 chills compared with version B, Wilcoxon Z(1) =−2.11, p < .05.
A similar analysis was performed for the remainingthree pieces that were repeated in the same version twotimes. These pieces were the Tuba Mirum (professionalrecording), the Bach motet (recording by Choir A), andthe Puccini Requiem by Puccini in a version (performedby choir B). No significant differences were found forthe first and second listening for any subgroup. The chillfrequency differed between groups for the first listeningof both the motet by Bach, (Kruskal-Wallis test, χ2(95) =8.36, p < .05) and Puccini’s Requiem, (Kruskal-Wallistest, χ2(95) = 10.91, p < .05). For both pieces the high-est numbers of chills was reported by the choir whosung that version.
Influence of Experience with Classical Music
To study the influence of general experience withclassical music, two groups were compared: Onegroup contained the chill responders (N = 15) whoreported to be very experienced with classical music(ratings of six and higher on a 7-point scale, upperquartile). The other group contained the chill respon-ders who reported to be unfamiliar with classical
music (ratings of two and lower on a 7-point scale,lower quartile).
The average of intensity reports and physiologicalresponses were compared (see Figure 5). While therewere no significant differences in SCR and HR (data forHR not shown), the ratings of intensity of feelings dif-fered significantly by approximately two units on a 20units scale. However, the course of the rating curves isvery similar, r(28) = .98, p < .001.
Active and Passive Listening to the Dies Irae
Comparing the active and the passive condition of theDies Irae, no significant differences in the frequency ofchills could be stated, neither in comparing all partici-pants (chills active: M = 1.19, SD = 3.48; chills passive:M = 0.89, SD = 1.89, Wilcoxon Z(1) = −3.38, n.s.), norwhen subdividing the group in test and control group.However, a difference in valence and arousal ratingscould be found, both being rated higher for the activecondition (valence active: M = 5.33, SD = 1.63, median =6, upper quartile = 7, lower quartile = 4, N = 93; valencepassive: M = 5.12, SD = 1.69, median = 6, upper quar-tile = 7, lower quartile = 4, N = 95, Wilcoxon Z(1) =−2.32, p < .05; arousal active: M = 6.26, SD = 1.06,median = 7, upper quartile = 7, lower quartile = 6, N =94; arousal passive: M = 5.99, SD = 1.07, median = 6,upper quartile = 7, lower quartile = 5, N = 95, WilcoxonZ(1) = −2.36, p < .05).
Discussion
Chills are Independent of Age, Gender,and Music Education
No significant differences could be found betweenfemale and male chill frequency. This confirms the
70 Oliver Grewe, Reinhard Kopiez, & Eckart Altenmüller
TABLE 4. Comparison of Averaged Number of Chills for Repeated Stimuli.
Choir C (N = 21) 0.95f 0.57 f 0.70 0.52 0.38 0.48Control group (N = 41) 0.35 0.39 0.34 0.39 0.15 0.27
↑ ↑χ2(3, N = 95) = 8.36 10.91
p < .05 p < .05
Note: f = familiar with piece, o = own recording (printed in bold).
Music2701_05 8/13/09 2:56 PM Page 70
findings of Goldstein (1980), and contradicts those byPanksepp (1995). In Panksepp’s study, participants lis-tened to the music in a group situation, whereas inGoldstein’s and in this study, participants listened tomusic in individual sessions. Thus, a social influencefactor may be hypothesized: females may tend to reportmore chills in a group situation. This finding supportsthe performing experiments in individual sessionswhen chills should be used as a parameter independentof gender. Furthermore, no relation between chill fre-quency and age were found. Similarly, music educationin general does not seem to play an important role inproducing chills. However, participants’ familiaritywith and liking of a piece of music are crucial for deter-mining the frequency of chill reactions. This is in accor-dance with the findings of Panksepp (1995). Here itshould be taken into account, of course, that partici-pants with higher levels of music education tend to bemore familiar with classical music. However, extensivemusic education does not mean the listener is familiarwith every piece of music. The topic of familiarity willbe addressed in more detail below.
The Subjective and Physiological Intensity of Chills
The comparison of chill samples and equivalent ran-dom non-chill samples over 61 chill responders revealedthat chills are perceived as more intense and are, at thesame time, reflected in a measurable physiologicalarousal response, namely the skin conductance andheart rate. An influence of breathing on the physiologi-cal results could be excluded. A steady increase in sub-jective intensity can be stated several seconds before the
chill is reported, culminating in a significant difference1.5 s before the chill onset. There was no further increasein intensity of feelings reported after the chill onset.These findings suggest that chills result from a cognitiveappraisal of the music and that the chills represent theclimax of a rising intensity of feelings as hypothesizedby Blood and Zatorre (2001). This interpretation wouldbe also in accordance with a previously published exper-iment (Grewe et al., 2007). The finding that chill-elicit-ing pieces were rated higher in valence and arousalallows an interpretation of the emotional quality of thereported intensity of feelings. This would simply meanthat chills are a pleasant and arousing peak, as assumedby all researchers who used them as a parameter so far.
Skin conductance and HR started rising 2 s precedingthe chill. Regarding the contradictory findings of Craig(2005) and Blood and Zatorre (2001) regarding skinconductance during chills, our results here confirm theresults of Craig. The increase in HR is in accordancewith the findings of Blood and Zatorre (2001) andSammler, Grigutsch, Fritz, and Koelsch (2007). Thus, itcan be confirmed that chills involve two of three emo-tion components; that is, subjective feeling and physio-logical arousal.
Interestingly, our findings fit the ITPRA Modelrecently suggested by Huron (2006). Huron suggestedthat chills are related to the surprise accompanying anevent. In a previous experiment, we showed a relationbetween chills and (a) the entrance of a voice or choir;and (b) the beginning of something new in the music(Grewe et al., 2007). The ITPRA model hypothesizesthat in anticipation of an event, the imagination (I) ofpossible outcomes of the event starts. In our study, this
The Chill Parameter 71
FIGURE 5. Comparison of chill samples (20 s) of participants highly experienced (arrowhead) and inexperienced (arrow) in classical music. Leftintensity of feeling ratings, right skin conductance response (SCR). Significant differences (permutation test, p < .05) are shaded grey.
Music2701_05 8/13/09 2:56 PM Page 71
would be analogous to the early increase of reportedsubjective intensity. Directly before the onset of theevent, tension (T) rises. This assumption would beanalogous to the rise of SCR and HR in our experi-ment. Following the event, the model expects an imme-diate reaction to the success or failure of the prediction(P), as well as a fast cursory adaptive reaction (R).These may be reflected by a second increase in SCRabout one second after the chill onset. Similar physio-logical correlates of musical expectancy have beenreported by other studies (Koelsch, Kilches, Steinbeis, &Schelinksi, 2008; Steinbeis, Koelsch, & Sloboda, 2006).Finally, feelings may be evoked as a conscious appraisal(A) of the event. Indeed in our study, after the chillonset, the reported feeling stayed on the high level ofintensity of feeling. These similarities between ourfindings and the suggested ITPRA model should, ofcourse, be tested in further experiments; however, theymay confirm the relevance of Huron’s model.
Effects of Familiarity on the Frequency of Chills
The experimental group contained more chill respon-ders than non-chill responders (72% vs. 56%).Participants from the experimental group also reportedthe most chills in total (679 vs. 173 chills), which meansthat 80% of the chills were reported by the experimen-tal group. A more detailed analysis was done by divid-ing the experimental group into three furthersubgroups: choir A, B, and C. The results can besummed up in three levels of familiarity for the pieces:unknown piece, familiar piece, and personal recordingof a piece. The highest level of familiarity does not seemto be of crucial importance for producing chills. Themean chill reactions for choirs A and B differed signifi-cantly in only one case in response to versions A and Bof the Lacrimosa, Confutatis, and Rex Tremendae. ChoirA reported significantly more chills for its own versionof the “Confutatis.” For the “Confutatis” and the “RexTremendae” a significant difference between the fourgroups was found. The control group reported lesschills compared to the participants familiar with thepieces. Generally this was true for all pieces.
Familiarity does not mean, however, that listenershave to be experienced listeners of classical music.Participants who were unfamiliar with classical musicreported a significantly lower level of subjective inten-sity during chill episodes. Regardless of the absolutelevel, though, intensity of feeling increased by approxi-mately 1.5 units in both experienced and inexperiencedlisteners. The physiological response confirmed thatchills show the same increase in physiological arousal
for experienced and inexperienced listeners. Up untilnow chills have been thought of as reactions related todistinct musical structures (Guhn et al., 2007; Sloboda,1991). The results presented here, as well as in a formerstudy (Grewe et al., 2007), indicate that emotionalpeaks might not be triggered in a simple stimulus-response manner and related to distinct musical struc-tures. Individual factors such as personal associationsmight be even more important. The most importantprecondition for using chills as a valuable research toolis a preselection of stimuli for a distinct group of par-ticipants. Researchers have to select a stimulus suitablefor the group they want to test, which means that par-ticipants generally should be familiar with the stimulusand like it. It would be an interesting task for the futureto test whether chills can be trained by introducing par-ticipants to a stimulus. If it is true that strong emotionalresponses also include cognitive components (Sander,Grandjean, & Scherer, 2005), then it is plausible thatchills cannot be simply triggered. If chills can reliablyindicate episodes of strong subjective feeling and phys-iological arousal, this makes them a promising indica-tor of strong emotion.
The Tuba Mirum, Bach motet, and Puccini Requiemwere played twice to all participants in the same ver-sion. On average, none of the groups reported a signif-icant difference of chills between the first and secondlistening. Between the groups a difference was foundfor the first listening of the Bach motet and PucciniRequiem. Interestingly, it was always the group familiarwith the piece that reported the most chills (i.e., choirA for the motet, and choir B for the Requiem). In thesecond listening session, the chills of the group familiarwith the piece decreased, while the chills of the othergroups slightly increased in tendency. Conclusionsbased on this rather exploratory part of the studyshould be drawn with care. However, the results suggestthat only a general familiarity with the stimulus has animportant effect on the chill frequency. Chills could notbe significantly increased by a more intimate knowl-edge of the stimulus. The repetition of the “TubaMirum,” the Bach motet and Puccini’s Requiemdemonstrated that chill responses can be stable for asecond listening and that the difference between famil-iar and non-familiar listeners diminishes even after thefirst listening of a piece.
Interestingly, a general familiarity with classicalmusic (independent of the piece being heard) led to adifference in the subjective perception of intensity ofchills; however, it did not alter the physiological arousalresponse. It also did not alter the finding that therewas an up-shift in the intensity of feelings. It could be
72 Oliver Grewe, Reinhard Kopiez, & Eckart Altenmüller
Music2701_05 8/13/09 2:56 PM Page 72
concluded that the stimulation of the peripheral nerv-ous system, perceived as a chill, is based on the dynam-ics in feelings rather than their absolute level. Thesudden evaluation process (e-motion) might be the rea-son for shiver and goose bump reactions. Here wemight also conclude that chills are not so much relatedto a distinct event, such as a distinct harmonic change.Chills might have very individual triggers; however, theunderlying cause might be a sudden and relevantchange in the stimulus, which leads to a strong dynamicin emotional evaluation. An increase in loudnessmight be the most prominent example. However, lis-teners turned out to be most individual in their capac-ity to detect musical changes. A listener who enjoysplaying the violoncello, for example, might perceivemusical events in the low strings more accurate thanothers.
Influence of Singing Along on Chills
It was shown that chills include two emotion compo-nents and we sought to control for possible influencesof the third component (motor responses) on chills.Neither an increase nor a decrease in frequency of chillscould be found due to active or passive listening. Due tothe preliminary character of this experiment, this maynot be a definite finding. For example, it would beinteresting to test the chill frequency while participantsdance to a piece, because dance may represent a morenatural response to music. In our laboratory setting,motor response showed no strong impact on chills.This means that it is not necessary to control how exactlyparticipants listen to the piece: whether they lip sync,whether they identify with the recorded performer, orwhether they imagine themselves as a passively lis-tening audience. The rather subtle comparisons drawnfrom the experimental conditions regarding theactive/passive as well as in the personal vs. other record-ing also can demonstrate the stability of chills towardssuch subtle (and hard to control) differences.
Conclusion
We conclude that chills can be a highly valuable instru-ment for emotion research, combining two of threebasal components. They present a suitable tool forexperiments on virtually all populations. However, dif-ficulties in using the chill parameter remain because ofthe high individual variance between responses. Theresults presented here may help make chills a systematicand regular tool not only in music-related emotionresearch: Chills have been reported also in otherdomains (Goldstein, 1980). The major advantage ofchills may be that they indicate individually occurringpeaks in feelings. If strong feelings involve cognitiveevaluation (Grewe et al., 2007; Scherer, 2004), theycould not be triggered simply at the same point in timefor all participants. Several participants might react tothe same stimulus; however their strongest reactionsmight be at different points in time. According to theresults presented here, chills can be interpreted as anindividual marker of emotional climax in subjectivefeeling as well as physiological arousal. If we succeed inbetter understanding the principles of this experience,the measurement of chills offers an additional valuablemethod for aesthetic psychology and emotion research.
Author Note
This work was supported by the DFG (Al 269-6) and theCentre for Systemic Neurosciences, Hannover. We alsowould like to acknowledge the work of our traineesBenedikt Zöfel, Kristina Schmidt, and Björn H. Katzur. Weapplied the FLAE approach for the sequence of authors.EMuJoy-software can be tested at: http://musicweb.hmt-hannover.de/emujoy/
Correspondence concerning this article should beaddressed to Eckart Altenmüller, Institute of MusicPhysiology and Musicians’ Medicine, Hannover Uni-versity of Music and Drama, Hohenzollernstr. 47 30161Hannover. E-MAIL: [email protected]
The Chill Parameter 73
References
BARDO, M. T. (1998). Neuropharmacological mechanisms of
drug reward: Beyond dopamine in the nucleus accumbens.
Critical Reviews in Neurobiology, 12, 37-67.
BERNTSON, G. G., CACIOPPO, J. T., & QUIGLEY, K. S. (1993).
Respiratory sinus arrhythmia: Autonomic origins,
physiological mechanisms, and psychophysiological
implications. Psychophysiology, 30, 183-196.
BLOOD, A. J., & ZATORRE, R. J. (2001). Intensely pleasurable
responses to music correlate with activity in brain regions
implicated in reward and emotion. Proceedings of the National
Academy of Sciences, 98, 11818-11823.
BOUCSEIN, W. (2001). Physiologische Grundlagen und Meßmeth-
oden der dermalen Aktivität [Physiological principles and
methods of measurements of dermal (skin) activity].
In F. Rösler (Ed.), Enzyklopädie der Psychologie, Bereich
Psychophysiologie [Cyclopedia of psychology, section
psychophysiology] (Vol. 1, pp. 551-623). Göttingen:
Hogrefe.
Music2701_05 8/13/09 2:56 PM Page 73
BREITER, H. C., GOLLUB, R. L., WEISSKOFF, R. M., KENNEDY,
D. N., MAKRIS, N., BERKE, J. D., ET AL. (1997). Acute
effects of cocaine on human brain activity and emotion.
Neuron, 19, 591-611.
CACIOPPO, J. T., KLEIN, D. J., BERNTSON, G. G., & HATFIELD,
E. (1993). The psychophysiology of emotion. New York:
Guilford.
CRAIG, D. G. (2005). An exploratory study of physiological
changes during “chills” induced by music. Musicae Scientiae,
IX, 273-287.
ELLSWORTH, P. C., & SCHERER, K. R. (2003). Appraisal
processes in emotion. In R. Davidson, K. R. Scherer, & H. H.
Goldsmith (Eds.), Handbook of the affective sciences (pp. 572-
596). New York: Oxford University Press
FRIJDA, N. H., & SUNDARARAJAN, L. (2007). Emotion
refinement. A theory inspired by Chinese poetics. Pespectives
on Psychological Science, 2, 227-241.
GOLDSTEIN, A. (1980). Thrills in response to music and other
stimuli. Physiological Psychology, 8, 126-129.
GOOD, P. (1994). Permutation tests: A practical guide to
resampling methods for testing hypotheses. New York: Springer.
GREWE, O., NAGEL, F., KOPIEZ, R., & ALTENMÜLLER, E.
(2007). Listening to music as a re-creative process—Physio-
logical, psychological and psychoacoustical correlates of chills
and strong emotions. Music Perception, 24, 297-314.
GUHN, M., HAMM, A., & ZENTNER, M. R. (2007).
Physiological and musico-acoustic correlates of the chill
response. Music Perception, 24, 473-483.
HURON, D. (2006). Sweet anticipation: Music and the psychology
of expectation. Cambridge, MA: A Bradford Book.
KARAJAN, H. V. (1989). “Tuba mirum” from Requiem KV626.
On Mozart-Requiem [CD]. Berlin: Deutsche Grammophon
(Universal).
KOELSCH, S., KILCHES, S., STEINBEIS, N., & SCHELINKSI, S.
(2008). Effects of unexpected chords and of performer’s
expression on brain responses and electrodermal activity.
